1
# Copyright (C) 2005, 2006, 2007 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""Knit versionedfile implementation.
19
A knit is a versioned file implementation that supports efficient append only
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
end-marker; simply "end VERSION"
28
delta can be line or full contents.a
29
... the 8's there are the index number of the annotation.
30
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
8 e.set('executable', 'yes')
36
8 if elt.get('executable') == 'yes':
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
43
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
44
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
45
09:33 < lifeless> right
46
09:33 < jrydberg> lifeless: the position and size is the range in the data file
49
so the index sequence is the dictionary compressed sequence number used
50
in the deltas to provide line annotation
55
# 10:16 < lifeless> make partial index writes safe
56
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
57
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
59
# move sha1 out of the content so that join is faster at verifying parents
60
# record content length ?
64
from cStringIO import StringIO
65
from itertools import izip, chain
70
from zlib import Z_DEFAULT_COMPRESSION
73
from bzrlib.lazy_import import lazy_import
74
lazy_import(globals(), """
94
from bzrlib.errors import (
102
RevisionAlreadyPresent,
104
from bzrlib.graph import Graph
105
from bzrlib.osutils import (
111
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
112
from bzrlib.tsort import topo_sort
113
from bzrlib.tuned_gzip import GzipFile, bytes_to_gzip
115
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
119
# TODO: Split out code specific to this format into an associated object.
121
# TODO: Can we put in some kind of value to check that the index and data
122
# files belong together?
124
# TODO: accommodate binaries, perhaps by storing a byte count
126
# TODO: function to check whole file
128
# TODO: atomically append data, then measure backwards from the cursor
129
# position after writing to work out where it was located. we may need to
130
# bypass python file buffering.
132
DATA_SUFFIX = '.knit'
133
INDEX_SUFFIX = '.kndx'
136
class KnitContent(object):
137
"""Content of a knit version to which deltas can be applied."""
140
self._should_strip_eol = False
143
"""Return a list of (origin, text) tuples."""
144
return list(self.annotate_iter())
146
def apply_delta(self, delta, new_version_id):
147
"""Apply delta to this object to become new_version_id."""
148
raise NotImplementedError(self.apply_delta)
150
def cleanup_eol(self, copy_on_mutate=True):
151
if self._should_strip_eol:
153
self._lines = self._lines[:]
154
self.strip_last_line_newline()
156
def line_delta_iter(self, new_lines):
157
"""Generate line-based delta from this content to new_lines."""
158
new_texts = new_lines.text()
159
old_texts = self.text()
160
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
161
for tag, i1, i2, j1, j2 in s.get_opcodes():
164
# ofrom, oto, length, data
165
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
167
def line_delta(self, new_lines):
168
return list(self.line_delta_iter(new_lines))
171
def get_line_delta_blocks(knit_delta, source, target):
172
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
173
target_len = len(target)
176
for s_begin, s_end, t_len, new_text in knit_delta:
177
true_n = s_begin - s_pos
180
# knit deltas do not provide reliable info about whether the
181
# last line of a file matches, due to eol handling.
182
if source[s_pos + n -1] != target[t_pos + n -1]:
185
yield s_pos, t_pos, n
186
t_pos += t_len + true_n
188
n = target_len - t_pos
190
if source[s_pos + n -1] != target[t_pos + n -1]:
193
yield s_pos, t_pos, n
194
yield s_pos + (target_len - t_pos), target_len, 0
197
class AnnotatedKnitContent(KnitContent):
198
"""Annotated content."""
200
def __init__(self, lines):
201
KnitContent.__init__(self)
204
def annotate_iter(self):
205
"""Yield tuples of (origin, text) for each content line."""
206
return iter(self._lines)
208
def apply_delta(self, delta, new_version_id):
209
"""Apply delta to this object to become new_version_id."""
212
for start, end, count, delta_lines in delta:
213
lines[offset+start:offset+end] = delta_lines
214
offset = offset + (start - end) + count
216
def strip_last_line_newline(self):
217
line = self._lines[-1][1].rstrip('\n')
218
self._lines[-1] = (self._lines[-1][0], line)
219
self._should_strip_eol = False
223
lines = [text for origin, text in self._lines]
224
except ValueError, e:
225
# most commonly (only?) caused by the internal form of the knit
226
# missing annotation information because of a bug - see thread
228
raise KnitCorrupt(self,
229
"line in annotated knit missing annotation information: %s"
232
if self._should_strip_eol:
233
anno, line = lines[-1]
234
lines[-1] = (anno, line.rstrip('\n'))
238
return AnnotatedKnitContent(self._lines[:])
241
class PlainKnitContent(KnitContent):
242
"""Unannotated content.
244
When annotate[_iter] is called on this content, the same version is reported
245
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
249
def __init__(self, lines, version_id):
250
KnitContent.__init__(self)
252
self._version_id = version_id
254
def annotate_iter(self):
255
"""Yield tuples of (origin, text) for each content line."""
256
for line in self._lines:
257
yield self._version_id, line
259
def apply_delta(self, delta, new_version_id):
260
"""Apply delta to this object to become new_version_id."""
263
for start, end, count, delta_lines in delta:
264
lines[offset+start:offset+end] = delta_lines
265
offset = offset + (start - end) + count
266
self._version_id = new_version_id
269
return PlainKnitContent(self._lines[:], self._version_id)
271
def strip_last_line_newline(self):
272
self._lines[-1] = self._lines[-1].rstrip('\n')
273
self._should_strip_eol = False
277
if self._should_strip_eol:
279
lines[-1] = lines[-1].rstrip('\n')
283
class _KnitFactory(object):
284
"""Base class for common Factory functions."""
286
def parse_record(self, version_id, record, record_details,
287
base_content, copy_base_content=True):
288
"""Parse a record into a full content object.
290
:param version_id: The official version id for this content
291
:param record: The data returned by read_records_iter()
292
:param record_details: Details about the record returned by
294
:param base_content: If get_build_details returns a compression_parent,
295
you must return a base_content here, else use None
296
:param copy_base_content: When building from the base_content, decide
297
you can either copy it and return a new object, or modify it in
299
:return: (content, delta) A Content object and possibly a line-delta,
302
method, noeol = record_details
303
if method == 'line-delta':
304
assert base_content is not None
305
if copy_base_content:
306
content = base_content.copy()
308
content = base_content
309
delta = self.parse_line_delta(record, version_id)
310
content.apply_delta(delta, version_id)
312
content = self.parse_fulltext(record, version_id)
314
content._should_strip_eol = noeol
315
return (content, delta)
318
class KnitAnnotateFactory(_KnitFactory):
319
"""Factory for creating annotated Content objects."""
323
def make(self, lines, version_id):
324
num_lines = len(lines)
325
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
327
def parse_fulltext(self, content, version_id):
328
"""Convert fulltext to internal representation
330
fulltext content is of the format
331
revid(utf8) plaintext\n
332
internal representation is of the format:
335
# TODO: jam 20070209 The tests expect this to be returned as tuples,
336
# but the code itself doesn't really depend on that.
337
# Figure out a way to not require the overhead of turning the
338
# list back into tuples.
339
lines = [tuple(line.split(' ', 1)) for line in content]
340
return AnnotatedKnitContent(lines)
342
def parse_line_delta_iter(self, lines):
343
return iter(self.parse_line_delta(lines))
345
def parse_line_delta(self, lines, version_id, plain=False):
346
"""Convert a line based delta into internal representation.
348
line delta is in the form of:
349
intstart intend intcount
351
revid(utf8) newline\n
352
internal representation is
353
(start, end, count, [1..count tuples (revid, newline)])
355
:param plain: If True, the lines are returned as a plain
356
list without annotations, not as a list of (origin, content) tuples, i.e.
357
(start, end, count, [1..count newline])
364
def cache_and_return(line):
365
origin, text = line.split(' ', 1)
366
return cache.setdefault(origin, origin), text
368
# walk through the lines parsing.
369
# Note that the plain test is explicitly pulled out of the
370
# loop to minimise any performance impact
373
start, end, count = [int(n) for n in header.split(',')]
374
contents = [next().split(' ', 1)[1] for i in xrange(count)]
375
result.append((start, end, count, contents))
378
start, end, count = [int(n) for n in header.split(',')]
379
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
380
result.append((start, end, count, contents))
383
def get_fulltext_content(self, lines):
384
"""Extract just the content lines from a fulltext."""
385
return (line.split(' ', 1)[1] for line in lines)
387
def get_linedelta_content(self, lines):
388
"""Extract just the content from a line delta.
390
This doesn't return all of the extra information stored in a delta.
391
Only the actual content lines.
396
header = header.split(',')
397
count = int(header[2])
398
for i in xrange(count):
399
origin, text = next().split(' ', 1)
402
def lower_fulltext(self, content):
403
"""convert a fulltext content record into a serializable form.
405
see parse_fulltext which this inverts.
407
# TODO: jam 20070209 We only do the caching thing to make sure that
408
# the origin is a valid utf-8 line, eventually we could remove it
409
return ['%s %s' % (o, t) for o, t in content._lines]
411
def lower_line_delta(self, delta):
412
"""convert a delta into a serializable form.
414
See parse_line_delta which this inverts.
416
# TODO: jam 20070209 We only do the caching thing to make sure that
417
# the origin is a valid utf-8 line, eventually we could remove it
419
for start, end, c, lines in delta:
420
out.append('%d,%d,%d\n' % (start, end, c))
421
out.extend(origin + ' ' + text
422
for origin, text in lines)
425
def annotate_iter(self, knit, version_id):
426
content = knit._get_content(version_id)
427
return content.annotate_iter()
430
class KnitPlainFactory(_KnitFactory):
431
"""Factory for creating plain Content objects."""
435
def make(self, lines, version_id):
436
return PlainKnitContent(lines, version_id)
438
def parse_fulltext(self, content, version_id):
439
"""This parses an unannotated fulltext.
441
Note that this is not a noop - the internal representation
442
has (versionid, line) - its just a constant versionid.
444
return self.make(content, version_id)
446
def parse_line_delta_iter(self, lines, version_id):
448
num_lines = len(lines)
449
while cur < num_lines:
452
start, end, c = [int(n) for n in header.split(',')]
453
yield start, end, c, lines[cur:cur+c]
456
def parse_line_delta(self, lines, version_id):
457
return list(self.parse_line_delta_iter(lines, version_id))
459
def get_fulltext_content(self, lines):
460
"""Extract just the content lines from a fulltext."""
463
def get_linedelta_content(self, lines):
464
"""Extract just the content from a line delta.
466
This doesn't return all of the extra information stored in a delta.
467
Only the actual content lines.
472
header = header.split(',')
473
count = int(header[2])
474
for i in xrange(count):
477
def lower_fulltext(self, content):
478
return content.text()
480
def lower_line_delta(self, delta):
482
for start, end, c, lines in delta:
483
out.append('%d,%d,%d\n' % (start, end, c))
487
def annotate_iter(self, knit, version_id):
488
annotator = _KnitAnnotator(knit)
489
return iter(annotator.annotate(version_id))
492
def make_empty_knit(transport, relpath):
493
"""Construct a empty knit at the specified location."""
494
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
497
class KnitVersionedFile(VersionedFile):
498
"""Weave-like structure with faster random access.
500
A knit stores a number of texts and a summary of the relationships
501
between them. Texts are identified by a string version-id. Texts
502
are normally stored and retrieved as a series of lines, but can
503
also be passed as single strings.
505
Lines are stored with the trailing newline (if any) included, to
506
avoid special cases for files with no final newline. Lines are
507
composed of 8-bit characters, not unicode. The combination of
508
these approaches should mean any 'binary' file can be safely
509
stored and retrieved.
512
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
513
factory=None, delta=True, create=False, create_parent_dir=False,
514
delay_create=False, dir_mode=None, index=None, access_method=None):
515
"""Construct a knit at location specified by relpath.
517
:param create: If not True, only open an existing knit.
518
:param create_parent_dir: If True, create the parent directory if
519
creating the file fails. (This is used for stores with
520
hash-prefixes that may not exist yet)
521
:param delay_create: The calling code is aware that the knit won't
522
actually be created until the first data is stored.
523
:param index: An index to use for the knit.
525
if access_mode is None:
527
super(KnitVersionedFile, self).__init__(access_mode)
528
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
529
self.transport = transport
530
self.filename = relpath
531
self.factory = factory or KnitAnnotateFactory()
532
self.writable = (access_mode == 'w')
535
self._max_delta_chain = 200
538
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
539
access_mode, create=create, file_mode=file_mode,
540
create_parent_dir=create_parent_dir, delay_create=delay_create,
544
if access_method is None:
545
_access = _KnitAccess(transport, relpath + DATA_SUFFIX, file_mode, dir_mode,
546
((create and not len(self)) and delay_create), create_parent_dir)
548
_access = access_method
549
if create and not len(self) and not delay_create:
551
self._data = _KnitData(_access)
554
return '%s(%s)' % (self.__class__.__name__,
555
self.transport.abspath(self.filename))
557
def _check_should_delta(self, first_parents):
558
"""Iterate back through the parent listing, looking for a fulltext.
560
This is used when we want to decide whether to add a delta or a new
561
fulltext. It searches for _max_delta_chain parents. When it finds a
562
fulltext parent, it sees if the total size of the deltas leading up to
563
it is large enough to indicate that we want a new full text anyway.
565
Return True if we should create a new delta, False if we should use a
570
delta_parents = first_parents
571
for count in xrange(self._max_delta_chain):
572
parent = delta_parents[0]
573
method = self._index.get_method(parent)
574
index, pos, size = self._index.get_position(parent)
575
if method == 'fulltext':
579
delta_parents = self._index.get_parent_map([parent])[parent]
581
# We couldn't find a fulltext, so we must create a new one
584
return fulltext_size > delta_size
586
def _add_raw_records(self, records, data):
587
"""Add all the records 'records' with data pre-joined in 'data'.
589
:param records: A list of tuples(version_id, options, parents, size).
590
:param data: The data for the records. When it is written, the records
591
are adjusted to have pos pointing into data by the sum of
592
the preceding records sizes.
595
raw_record_sizes = [record[3] for record in records]
596
positions = self._data.add_raw_records(raw_record_sizes, data)
599
for (version_id, options, parents, size), access_memo in zip(
601
index_entries.append((version_id, options, access_memo, parents))
602
if self._data._do_cache:
603
self._data._cache[version_id] = data[offset:offset+size]
605
self._index.add_versions(index_entries)
607
def enable_cache(self):
608
"""Start caching data for this knit"""
609
self._data.enable_cache()
611
def clear_cache(self):
612
"""Clear the data cache only."""
613
self._data.clear_cache()
615
def copy_to(self, name, transport):
616
"""See VersionedFile.copy_to()."""
617
# copy the current index to a temp index to avoid racing with local
619
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
620
self.transport.get(self._index._filename))
622
f = self._data._open_file()
624
transport.put_file(name + DATA_SUFFIX, f)
627
# move the copied index into place
628
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
630
def create_empty(self, name, transport, mode=None):
631
return KnitVersionedFile(name, transport, factory=self.factory,
632
delta=self.delta, create=True)
634
def get_data_stream(self, required_versions):
635
"""Get a data stream for the specified versions.
637
Versions may be returned in any order, not necessarily the order
638
specified. They are returned in a partial order by compression
639
parent, so that the deltas can be applied as the data stream is
640
inserted; however note that compression parents will not be sent
641
unless they were specifically requested, as the client may already
644
:param required_versions: The exact set of versions to be extracted.
645
Unlike some other knit methods, this is not used to generate a
646
transitive closure, rather it is used precisely as given.
648
:returns: format_signature, list of (version, options, length, parents),
651
required_version_set = frozenset(required_versions)
653
# list of revisions that can just be sent without waiting for their
656
# map from revision to the children based on it
658
# first, read all relevant index data, enough to sort into the right
660
for version_id in required_versions:
661
options = self._index.get_options(version_id)
662
parents = self._index.get_parents_with_ghosts(version_id)
663
index_memo = self._index.get_position(version_id)
664
version_index[version_id] = (index_memo, options, parents)
665
if ('line-delta' in options
666
and parents[0] in required_version_set):
667
# must wait until the parent has been sent
668
deferred.setdefault(parents[0], []). \
671
# either a fulltext, or a delta whose parent the client did
672
# not ask for and presumably already has
673
ready_to_send.append(version_id)
674
# build a list of results to return, plus instructions for data to
676
copy_queue_records = []
677
temp_version_list = []
679
# XXX: pushing and popping lists may be a bit inefficient
680
version_id = ready_to_send.pop(0)
681
(index_memo, options, parents) = version_index[version_id]
682
copy_queue_records.append((version_id, index_memo))
683
none, data_pos, data_size = index_memo
684
temp_version_list.append((version_id, options, data_size,
686
if version_id in deferred:
687
# now we can send all the children of this revision - we could
688
# put them in anywhere, but we hope that sending them soon
689
# after the fulltext will give good locality in the receiver
690
ready_to_send[:0] = deferred.pop(version_id)
691
assert len(deferred) == 0, \
692
"Still have compressed child versions waiting to be sent"
693
# XXX: The stream format is such that we cannot stream it - we have to
694
# know the length of all the data a-priori.
696
result_version_list = []
697
for (version_id, raw_data), \
698
(version_id2, options, _, parents) in \
699
izip(self._data.read_records_iter_raw(copy_queue_records),
701
assert version_id == version_id2, \
702
'logic error, inconsistent results'
703
raw_datum.append(raw_data)
704
result_version_list.append(
705
(version_id, options, len(raw_data), parents))
706
# provide a callback to get data incrementally.
707
pseudo_file = StringIO(''.join(raw_datum))
710
return pseudo_file.read()
712
return pseudo_file.read(length)
713
return (self.get_format_signature(), result_version_list, read)
715
def _extract_blocks(self, version_id, source, target):
716
if self._index.get_method(version_id) != 'line-delta':
718
parent, sha1, noeol, delta = self.get_delta(version_id)
719
return KnitContent.get_line_delta_blocks(delta, source, target)
721
def get_delta(self, version_id):
722
"""Get a delta for constructing version from some other version."""
723
self.check_not_reserved_id(version_id)
724
parents = self.get_parent_map([version_id])[version_id]
729
index_memo = self._index.get_position(version_id)
730
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
731
noeol = 'no-eol' in self._index.get_options(version_id)
732
if 'fulltext' == self._index.get_method(version_id):
733
new_content = self.factory.parse_fulltext(data, version_id)
734
if parent is not None:
735
reference_content = self._get_content(parent)
736
old_texts = reference_content.text()
739
new_texts = new_content.text()
740
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
742
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
744
delta = self.factory.parse_line_delta(data, version_id)
745
return parent, sha1, noeol, delta
747
def get_format_signature(self):
748
"""See VersionedFile.get_format_signature()."""
749
if self.factory.annotated:
750
annotated_part = "annotated"
752
annotated_part = "plain"
753
return "knit-%s" % (annotated_part,)
755
def get_graph_with_ghosts(self):
756
"""See VersionedFile.get_graph_with_ghosts()."""
757
graph_items = self._index.get_graph()
758
return dict(graph_items)
760
def get_sha1(self, version_id):
761
return self.get_sha1s([version_id])[0]
763
def get_sha1s(self, version_ids):
764
"""See VersionedFile.get_sha1()."""
765
record_map = self._get_record_map(version_ids)
766
# record entry 2 is the 'digest'.
767
return [record_map[v][2] for v in version_ids]
771
"""See VersionedFile.get_suffixes()."""
772
return [DATA_SUFFIX, INDEX_SUFFIX]
774
def has_ghost(self, version_id):
775
"""True if there is a ghost reference in the file to version_id."""
777
if self.has_version(version_id):
779
# optimisable if needed by memoising the _ghosts set.
780
items = self._index.get_graph()
781
for node, parents in items:
782
for parent in parents:
783
if parent not in self._index._cache:
784
if parent == version_id:
788
def insert_data_stream(self, (format, data_list, reader_callable)):
789
"""Insert knit records from a data stream into this knit.
791
If a version in the stream is already present in this knit, it will not
792
be inserted a second time. It will be checked for consistency with the
793
stored version however, and may cause a KnitCorrupt error to be raised
794
if the data in the stream disagrees with the already stored data.
796
:seealso: get_data_stream
798
if format != self.get_format_signature():
799
if 'knit' in debug.debug_flags:
801
'incompatible format signature inserting to %r', self)
802
source = self._knit_from_datastream(
803
(format, data_list, reader_callable))
807
for version_id, options, length, parents in data_list:
808
if self.has_version(version_id):
809
# First check: the list of parents.
810
my_parents = self.get_parents_with_ghosts(version_id)
811
if tuple(my_parents) != tuple(parents):
812
# XXX: KnitCorrupt is not quite the right exception here.
815
'parents list %r from data stream does not match '
816
'already recorded parents %r for %s'
817
% (parents, my_parents, version_id))
819
# Also check the SHA-1 of the fulltext this content will
821
raw_data = reader_callable(length)
822
my_fulltext_sha1 = self.get_sha1(version_id)
823
df, rec = self._data._parse_record_header(version_id, raw_data)
824
stream_fulltext_sha1 = rec[3]
825
if my_fulltext_sha1 != stream_fulltext_sha1:
826
# Actually, we don't know if it's this knit that's corrupt,
827
# or the data stream we're trying to insert.
829
self.filename, 'sha-1 does not match %s' % version_id)
831
if 'line-delta' in options:
832
# Make sure that this knit record is actually useful: a
833
# line-delta is no use unless we have its parent.
834
# Fetching from a broken repository with this problem
835
# shouldn't break the target repository.
837
# See https://bugs.launchpad.net/bzr/+bug/164443
838
if not self._index.has_version(parents[0]):
841
'line-delta from stream '
844
'missing parent %s\n'
845
'Try running "bzr check" '
846
'on the source repository, and "bzr reconcile" '
848
(version_id, parents[0]))
849
self._add_raw_records(
850
[(version_id, options, parents, length)],
851
reader_callable(length))
853
def _knit_from_datastream(self, (format, data_list, reader_callable)):
854
"""Create a knit object from a data stream.
856
This method exists to allow conversion of data streams that do not
857
match the signature of this knit. Generally it will be slower and use
858
more memory to use this method to insert data, but it will work.
860
:seealso: get_data_stream for details on datastreams.
861
:return: A knit versioned file which can be used to join the datastream
864
if format == "knit-plain":
865
factory = KnitPlainFactory()
866
elif format == "knit-annotated":
867
factory = KnitAnnotateFactory()
869
raise errors.KnitDataStreamUnknown(format)
870
index = _StreamIndex(data_list, self._index)
871
access = _StreamAccess(reader_callable, index, self, factory)
872
return KnitVersionedFile(self.filename, self.transport,
873
factory=factory, index=index, access_method=access)
876
"""See VersionedFile.versions."""
877
if 'evil' in debug.debug_flags:
878
trace.mutter_callsite(2, "versions scales with size of history")
879
return self._index.get_versions()
881
def has_version(self, version_id):
882
"""See VersionedFile.has_version."""
883
if 'evil' in debug.debug_flags:
884
trace.mutter_callsite(2, "has_version is a LBYL scenario")
885
return self._index.has_version(version_id)
887
__contains__ = has_version
889
def _merge_annotations(self, content, parents, parent_texts={},
890
delta=None, annotated=None,
891
left_matching_blocks=None):
892
"""Merge annotations for content. This is done by comparing
893
the annotations based on changed to the text.
895
if left_matching_blocks is not None:
896
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
900
for parent_id in parents:
901
merge_content = self._get_content(parent_id, parent_texts)
902
if (parent_id == parents[0] and delta_seq is not None):
905
seq = patiencediff.PatienceSequenceMatcher(
906
None, merge_content.text(), content.text())
907
for i, j, n in seq.get_matching_blocks():
910
# this appears to copy (origin, text) pairs across to the
911
# new content for any line that matches the last-checked
913
content._lines[j:j+n] = merge_content._lines[i:i+n]
915
if delta_seq is None:
916
reference_content = self._get_content(parents[0], parent_texts)
917
new_texts = content.text()
918
old_texts = reference_content.text()
919
delta_seq = patiencediff.PatienceSequenceMatcher(
920
None, old_texts, new_texts)
921
return self._make_line_delta(delta_seq, content)
923
def _make_line_delta(self, delta_seq, new_content):
924
"""Generate a line delta from delta_seq and new_content."""
926
for op in delta_seq.get_opcodes():
929
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
932
def _get_components_positions(self, version_ids):
933
"""Produce a map of position data for the components of versions.
935
This data is intended to be used for retrieving the knit records.
937
A dict of version_id to (record_details, index_memo, next, parents) is
939
method is the way referenced data should be applied.
940
index_memo is the handle to pass to the data access to actually get the
942
next is the build-parent of the version, or None for fulltexts.
943
parents is the version_ids of the parents of this version
946
pending_components = version_ids
947
while pending_components:
948
build_details = self._index.get_build_details(pending_components)
949
current_components = set(pending_components)
950
pending_components = set()
951
for version_id, details in build_details.iteritems():
952
(index_memo, compression_parent, parents,
953
record_details) = details
954
method = record_details[0]
955
if compression_parent is not None:
956
pending_components.add(compression_parent)
957
component_data[version_id] = (record_details, index_memo,
959
missing = current_components.difference(build_details)
961
raise errors.RevisionNotPresent(missing.pop(), self.filename)
962
return component_data
964
def _get_content(self, version_id, parent_texts={}):
965
"""Returns a content object that makes up the specified
967
cached_version = parent_texts.get(version_id, None)
968
if cached_version is not None:
969
if not self.has_version(version_id):
970
raise RevisionNotPresent(version_id, self.filename)
971
return cached_version
973
text_map, contents_map = self._get_content_maps([version_id])
974
return contents_map[version_id]
976
def _check_versions_present(self, version_ids):
977
"""Check that all specified versions are present."""
978
self._index.check_versions_present(version_ids)
980
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
981
nostore_sha, random_id, check_content, left_matching_blocks):
982
"""See VersionedFile.add_lines_with_ghosts()."""
983
self._check_add(version_id, lines, random_id, check_content)
984
return self._add(version_id, lines, parents, self.delta,
985
parent_texts, left_matching_blocks, nostore_sha, random_id)
987
def _add_lines(self, version_id, parents, lines, parent_texts,
988
left_matching_blocks, nostore_sha, random_id, check_content):
989
"""See VersionedFile.add_lines."""
990
self._check_add(version_id, lines, random_id, check_content)
991
self._check_versions_present(parents)
992
return self._add(version_id, lines[:], parents, self.delta,
993
parent_texts, left_matching_blocks, nostore_sha, random_id)
995
def _check_add(self, version_id, lines, random_id, check_content):
996
"""check that version_id and lines are safe to add."""
997
if contains_whitespace(version_id):
998
raise InvalidRevisionId(version_id, self.filename)
999
self.check_not_reserved_id(version_id)
1000
# Technically this could be avoided if we are happy to allow duplicate
1001
# id insertion when other things than bzr core insert texts, but it
1002
# seems useful for folk using the knit api directly to have some safety
1003
# blanket that we can disable.
1004
if not random_id and self.has_version(version_id):
1005
raise RevisionAlreadyPresent(version_id, self.filename)
1007
self._check_lines_not_unicode(lines)
1008
self._check_lines_are_lines(lines)
1010
def _add(self, version_id, lines, parents, delta, parent_texts,
1011
left_matching_blocks, nostore_sha, random_id):
1012
"""Add a set of lines on top of version specified by parents.
1014
If delta is true, compress the text as a line-delta against
1017
Any versions not present will be converted into ghosts.
1019
# first thing, if the content is something we don't need to store, find
1021
line_bytes = ''.join(lines)
1022
digest = sha_string(line_bytes)
1023
if nostore_sha == digest:
1024
raise errors.ExistingContent
1026
present_parents = []
1027
if parent_texts is None:
1029
for parent in parents:
1030
if self.has_version(parent):
1031
present_parents.append(parent)
1033
# can only compress against the left most present parent.
1035
(len(present_parents) == 0 or
1036
present_parents[0] != parents[0])):
1039
text_length = len(line_bytes)
1042
if lines[-1][-1] != '\n':
1043
# copy the contents of lines.
1045
options.append('no-eol')
1046
lines[-1] = lines[-1] + '\n'
1050
# To speed the extract of texts the delta chain is limited
1051
# to a fixed number of deltas. This should minimize both
1052
# I/O and the time spend applying deltas.
1053
delta = self._check_should_delta(present_parents)
1055
assert isinstance(version_id, str)
1056
content = self.factory.make(lines, version_id)
1057
if delta or (self.factory.annotated and len(present_parents) > 0):
1058
# Merge annotations from parent texts if needed.
1059
delta_hunks = self._merge_annotations(content, present_parents,
1060
parent_texts, delta, self.factory.annotated,
1061
left_matching_blocks)
1064
options.append('line-delta')
1065
store_lines = self.factory.lower_line_delta(delta_hunks)
1066
size, bytes = self._data._record_to_data(version_id, digest,
1069
options.append('fulltext')
1070
# isinstance is slower and we have no hierarchy.
1071
if self.factory.__class__ == KnitPlainFactory:
1072
# Use the already joined bytes saving iteration time in
1074
size, bytes = self._data._record_to_data(version_id, digest,
1075
lines, [line_bytes])
1077
# get mixed annotation + content and feed it into the
1079
store_lines = self.factory.lower_fulltext(content)
1080
size, bytes = self._data._record_to_data(version_id, digest,
1083
access_memo = self._data.add_raw_records([size], bytes)[0]
1084
self._index.add_versions(
1085
((version_id, options, access_memo, parents),),
1086
random_id=random_id)
1087
return digest, text_length, content
1089
def check(self, progress_bar=None):
1090
"""See VersionedFile.check()."""
1092
def _clone_text(self, new_version_id, old_version_id, parents):
1093
"""See VersionedFile.clone_text()."""
1094
# FIXME RBC 20060228 make fast by only inserting an index with null
1096
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
1098
def get_lines(self, version_id):
1099
"""See VersionedFile.get_lines()."""
1100
return self.get_line_list([version_id])[0]
1102
def _get_record_map(self, version_ids):
1103
"""Produce a dictionary of knit records.
1105
:return: {version_id:(record, record_details, digest, next)}
1107
data returned from read_records
1109
opaque information to pass to parse_record
1111
SHA1 digest of the full text after all steps are done
1113
build-parent of the version, i.e. the leftmost ancestor.
1114
Will be None if the record is not a delta.
1116
position_map = self._get_components_positions(version_ids)
1117
# c = component_id, r = record_details, i_m = index_memo, n = next
1118
records = [(c, i_m) for c, (r, i_m, n)
1119
in position_map.iteritems()]
1121
for component_id, record, digest in \
1122
self._data.read_records_iter(records):
1123
(record_details, index_memo, next) = position_map[component_id]
1124
record_map[component_id] = record, record_details, digest, next
1128
def get_text(self, version_id):
1129
"""See VersionedFile.get_text"""
1130
return self.get_texts([version_id])[0]
1132
def get_texts(self, version_ids):
1133
return [''.join(l) for l in self.get_line_list(version_ids)]
1135
def get_line_list(self, version_ids):
1136
"""Return the texts of listed versions as a list of strings."""
1137
for version_id in version_ids:
1138
self.check_not_reserved_id(version_id)
1139
text_map, content_map = self._get_content_maps(version_ids)
1140
return [text_map[v] for v in version_ids]
1142
_get_lf_split_line_list = get_line_list
1144
def _get_content_maps(self, version_ids):
1145
"""Produce maps of text and KnitContents
1147
:return: (text_map, content_map) where text_map contains the texts for
1148
the requested versions and content_map contains the KnitContents.
1149
Both dicts take version_ids as their keys.
1151
# FUTURE: This function could be improved for the 'extract many' case
1152
# by tracking each component and only doing the copy when the number of
1153
# children than need to apply delta's to it is > 1 or it is part of the
1155
version_ids = list(version_ids)
1156
multiple_versions = len(version_ids) != 1
1157
record_map = self._get_record_map(version_ids)
1162
for version_id in version_ids:
1165
while cursor is not None:
1166
record, record_details, digest, next = record_map[cursor]
1167
components.append((cursor, record, record_details, digest))
1168
if cursor in content_map:
1173
for (component_id, record, record_details,
1174
digest) in reversed(components):
1175
if component_id in content_map:
1176
content = content_map[component_id]
1178
content, delta = self.factory.parse_record(version_id,
1179
record, record_details, content,
1180
copy_base_content=multiple_versions)
1181
if multiple_versions:
1182
content_map[component_id] = content
1184
content.cleanup_eol(copy_on_mutate=multiple_versions)
1185
final_content[version_id] = content
1187
# digest here is the digest from the last applied component.
1188
text = content.text()
1189
actual_sha = sha_strings(text)
1190
if actual_sha != digest:
1191
raise KnitCorrupt(self.filename,
1193
'\n of reconstructed text does not match'
1195
'\n for version %s' %
1196
(actual_sha, digest, version_id))
1197
text_map[version_id] = text
1198
return text_map, final_content
1200
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1202
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1203
if version_ids is None:
1204
version_ids = self.versions()
1206
pb = progress.DummyProgress()
1207
# we don't care about inclusions, the caller cares.
1208
# but we need to setup a list of records to visit.
1209
# we need version_id, position, length
1210
version_id_records = []
1211
requested_versions = set(version_ids)
1212
# filter for available versions
1213
for version_id in requested_versions:
1214
if not self.has_version(version_id):
1215
raise RevisionNotPresent(version_id, self.filename)
1216
# get a in-component-order queue:
1217
for version_id in self.versions():
1218
if version_id in requested_versions:
1219
index_memo = self._index.get_position(version_id)
1220
version_id_records.append((version_id, index_memo))
1222
total = len(version_id_records)
1223
for version_idx, (version_id, data, sha_value) in \
1224
enumerate(self._data.read_records_iter(version_id_records)):
1225
pb.update('Walking content.', version_idx, total)
1226
method = self._index.get_method(version_id)
1228
assert method in ('fulltext', 'line-delta')
1229
if method == 'fulltext':
1230
line_iterator = self.factory.get_fulltext_content(data)
1232
line_iterator = self.factory.get_linedelta_content(data)
1233
# XXX: It might be more efficient to yield (version_id,
1234
# line_iterator) in the future. However for now, this is a simpler
1235
# change to integrate into the rest of the codebase. RBC 20071110
1236
for line in line_iterator:
1237
yield line, version_id
1239
pb.update('Walking content.', total, total)
1241
def iter_parents(self, version_ids):
1242
"""Iterate through the parents for many version ids.
1244
:param version_ids: An iterable yielding version_ids.
1245
:return: An iterator that yields (version_id, parents). Requested
1246
version_ids not present in the versioned file are simply skipped.
1247
The order is undefined, allowing for different optimisations in
1248
the underlying implementation.
1250
return self._index.iter_parents(version_ids)
1252
def num_versions(self):
1253
"""See VersionedFile.num_versions()."""
1254
return self._index.num_versions()
1256
__len__ = num_versions
1258
def annotate_iter(self, version_id):
1259
"""See VersionedFile.annotate_iter."""
1260
return self.factory.annotate_iter(self, version_id)
1262
def get_parent_map(self, version_ids):
1263
"""See VersionedFile.get_parent_map."""
1264
return self._index.get_parent_map(version_ids)
1266
def get_ancestry(self, versions, topo_sorted=True):
1267
"""See VersionedFile.get_ancestry."""
1268
if isinstance(versions, basestring):
1269
versions = [versions]
1272
return self._index.get_ancestry(versions, topo_sorted)
1274
def get_ancestry_with_ghosts(self, versions):
1275
"""See VersionedFile.get_ancestry_with_ghosts."""
1276
if isinstance(versions, basestring):
1277
versions = [versions]
1280
return self._index.get_ancestry_with_ghosts(versions)
1282
def plan_merge(self, ver_a, ver_b):
1283
"""See VersionedFile.plan_merge."""
1284
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1285
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1286
annotated_a = self.annotate(ver_a)
1287
annotated_b = self.annotate(ver_b)
1288
return merge._plan_annotate_merge(annotated_a, annotated_b,
1289
ancestors_a, ancestors_b)
1292
class _KnitComponentFile(object):
1293
"""One of the files used to implement a knit database"""
1295
def __init__(self, transport, filename, mode, file_mode=None,
1296
create_parent_dir=False, dir_mode=None):
1297
self._transport = transport
1298
self._filename = filename
1300
self._file_mode = file_mode
1301
self._dir_mode = dir_mode
1302
self._create_parent_dir = create_parent_dir
1303
self._need_to_create = False
1305
def _full_path(self):
1306
"""Return the full path to this file."""
1307
return self._transport.base + self._filename
1309
def check_header(self, fp):
1310
line = fp.readline()
1312
# An empty file can actually be treated as though the file doesn't
1314
raise errors.NoSuchFile(self._full_path())
1315
if line != self.HEADER:
1316
raise KnitHeaderError(badline=line,
1317
filename=self._transport.abspath(self._filename))
1320
return '%s(%s)' % (self.__class__.__name__, self._filename)
1323
class _KnitIndex(_KnitComponentFile):
1324
"""Manages knit index file.
1326
The index is already kept in memory and read on startup, to enable
1327
fast lookups of revision information. The cursor of the index
1328
file is always pointing to the end, making it easy to append
1331
_cache is a cache for fast mapping from version id to a Index
1334
_history is a cache for fast mapping from indexes to version ids.
1336
The index data format is dictionary compressed when it comes to
1337
parent references; a index entry may only have parents that with a
1338
lover index number. As a result, the index is topological sorted.
1340
Duplicate entries may be written to the index for a single version id
1341
if this is done then the latter one completely replaces the former:
1342
this allows updates to correct version and parent information.
1343
Note that the two entries may share the delta, and that successive
1344
annotations and references MUST point to the first entry.
1346
The index file on disc contains a header, followed by one line per knit
1347
record. The same revision can be present in an index file more than once.
1348
The first occurrence gets assigned a sequence number starting from 0.
1350
The format of a single line is
1351
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1352
REVISION_ID is a utf8-encoded revision id
1353
FLAGS is a comma separated list of flags about the record. Values include
1354
no-eol, line-delta, fulltext.
1355
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1356
that the the compressed data starts at.
1357
LENGTH is the ascii representation of the length of the data file.
1358
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1360
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1361
revision id already in the knit that is a parent of REVISION_ID.
1362
The ' :' marker is the end of record marker.
1365
when a write is interrupted to the index file, it will result in a line
1366
that does not end in ' :'. If the ' :' is not present at the end of a line,
1367
or at the end of the file, then the record that is missing it will be
1368
ignored by the parser.
1370
When writing new records to the index file, the data is preceded by '\n'
1371
to ensure that records always start on new lines even if the last write was
1372
interrupted. As a result its normal for the last line in the index to be
1373
missing a trailing newline. One can be added with no harmful effects.
1376
HEADER = "# bzr knit index 8\n"
1378
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1379
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1381
def _cache_version(self, version_id, options, pos, size, parents):
1382
"""Cache a version record in the history array and index cache.
1384
This is inlined into _load_data for performance. KEEP IN SYNC.
1385
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1388
# only want the _history index to reference the 1st index entry
1390
if version_id not in self._cache:
1391
index = len(self._history)
1392
self._history.append(version_id)
1394
index = self._cache[version_id][5]
1395
self._cache[version_id] = (version_id,
1402
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1403
create_parent_dir=False, delay_create=False, dir_mode=None):
1404
_KnitComponentFile.__init__(self, transport, filename, mode,
1405
file_mode=file_mode,
1406
create_parent_dir=create_parent_dir,
1409
# position in _history is the 'official' index for a revision
1410
# but the values may have come from a newer entry.
1411
# so - wc -l of a knit index is != the number of unique names
1415
fp = self._transport.get(self._filename)
1417
# _load_data may raise NoSuchFile if the target knit is
1419
_load_data(self, fp)
1423
if mode != 'w' or not create:
1426
self._need_to_create = True
1428
self._transport.put_bytes_non_atomic(
1429
self._filename, self.HEADER, mode=self._file_mode)
1431
def get_graph(self):
1432
"""Return a list of the node:parents lists from this knit index."""
1433
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1435
def get_ancestry(self, versions, topo_sorted=True):
1436
"""See VersionedFile.get_ancestry."""
1437
# get a graph of all the mentioned versions:
1439
pending = set(versions)
1442
version = pending.pop()
1445
parents = [p for p in cache[version][4] if p in cache]
1447
raise RevisionNotPresent(version, self._filename)
1448
# if not completed and not a ghost
1449
pending.update([p for p in parents if p not in graph])
1450
graph[version] = parents
1453
return topo_sort(graph.items())
1455
def get_ancestry_with_ghosts(self, versions):
1456
"""See VersionedFile.get_ancestry_with_ghosts."""
1457
# get a graph of all the mentioned versions:
1458
self.check_versions_present(versions)
1461
pending = set(versions)
1463
version = pending.pop()
1465
parents = cache[version][4]
1471
pending.update([p for p in parents if p not in graph])
1472
graph[version] = parents
1473
return topo_sort(graph.items())
1475
def get_build_details(self, version_ids):
1476
"""Get the method, index_memo and compression parent for version_ids.
1478
Ghosts are omitted from the result.
1480
:param version_ids: An iterable of version_ids.
1481
:return: A dict of version_id:(index_memo, compression_parent,
1482
parents, record_details).
1484
opaque structure to pass to read_records to extract the raw
1487
Content that this record is built upon, may be None
1489
Logical parents of this node
1491
extra information about the content which needs to be passed to
1492
Factory.parse_record
1495
for version_id in version_ids:
1496
if version_id not in self._cache:
1497
# ghosts are omitted
1499
method = self.get_method(version_id)
1500
parents = self.get_parents_with_ghosts(version_id)
1501
if method == 'fulltext':
1502
compression_parent = None
1504
compression_parent = parents[0]
1505
noeol = 'no-eol' in self.get_options(version_id)
1506
index_memo = self.get_position(version_id)
1507
result[version_id] = (index_memo, compression_parent,
1508
parents, (method, noeol))
1511
def iter_parents(self, version_ids):
1512
"""Iterate through the parents for many version ids.
1514
:param version_ids: An iterable yielding version_ids.
1515
:return: An iterator that yields (version_id, parents). Requested
1516
version_ids not present in the versioned file are simply skipped.
1517
The order is undefined, allowing for different optimisations in
1518
the underlying implementation.
1520
parent_map = self.get_parent_map(version_ids)
1521
parent_map_set = set(parent_map)
1522
unknown_existence = set()
1523
for parents in parent_map.itervalues():
1524
unknown_existence.update(parents)
1525
unknown_existence.difference_update(parent_map_set)
1526
present_parents = set(self.get_parent_map(unknown_existence))
1527
present_parents.update(parent_map_set)
1528
for version_id, parents in parent_map.iteritems():
1529
parents = tuple(parent for parent in parents
1530
if parent in present_parents)
1531
yield version_id, parents
1533
def num_versions(self):
1534
return len(self._history)
1536
__len__ = num_versions
1538
def get_versions(self):
1539
"""Get all the versions in the file. not topologically sorted."""
1540
return self._history
1542
def _version_list_to_index(self, versions):
1545
for version in versions:
1546
if version in cache:
1547
# -- inlined lookup() --
1548
result_list.append(str(cache[version][5]))
1549
# -- end lookup () --
1551
result_list.append('.' + version)
1552
return ' '.join(result_list)
1554
def add_version(self, version_id, options, index_memo, parents):
1555
"""Add a version record to the index."""
1556
self.add_versions(((version_id, options, index_memo, parents),))
1558
def add_versions(self, versions, random_id=False):
1559
"""Add multiple versions to the index.
1561
:param versions: a list of tuples:
1562
(version_id, options, pos, size, parents).
1563
:param random_id: If True the ids being added were randomly generated
1564
and no check for existence will be performed.
1567
orig_history = self._history[:]
1568
orig_cache = self._cache.copy()
1571
for version_id, options, (index, pos, size), parents in versions:
1572
line = "\n%s %s %s %s %s :" % (version_id,
1576
self._version_list_to_index(parents))
1577
assert isinstance(line, str), \
1578
'content must be utf-8 encoded: %r' % (line,)
1580
self._cache_version(version_id, options, pos, size, tuple(parents))
1581
if not self._need_to_create:
1582
self._transport.append_bytes(self._filename, ''.join(lines))
1585
sio.write(self.HEADER)
1586
sio.writelines(lines)
1588
self._transport.put_file_non_atomic(self._filename, sio,
1589
create_parent_dir=self._create_parent_dir,
1590
mode=self._file_mode,
1591
dir_mode=self._dir_mode)
1592
self._need_to_create = False
1594
# If any problems happen, restore the original values and re-raise
1595
self._history = orig_history
1596
self._cache = orig_cache
1599
def has_version(self, version_id):
1600
"""True if the version is in the index."""
1601
return version_id in self._cache
1603
def get_position(self, version_id):
1604
"""Return details needed to access the version.
1606
.kndx indices do not support split-out data, so return None for the
1609
:return: a tuple (None, data position, size) to hand to the access
1610
logic to get the record.
1612
entry = self._cache[version_id]
1613
return None, entry[2], entry[3]
1615
def get_method(self, version_id):
1616
"""Return compression method of specified version."""
1618
options = self._cache[version_id][1]
1620
raise RevisionNotPresent(version_id, self._filename)
1621
if 'fulltext' in options:
1624
if 'line-delta' not in options:
1625
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1628
def get_options(self, version_id):
1629
"""Return a list representing options.
1633
return self._cache[version_id][1]
1635
def get_parent_map(self, version_ids):
1636
"""Passed through to by KnitVersionedFile.get_parent_map."""
1638
for version_id in version_ids:
1640
result[version_id] = tuple(self._cache[version_id][4])
1645
def get_parents_with_ghosts(self, version_id):
1646
"""Return parents of specified version with ghosts."""
1648
return self.get_parent_map([version_id])[version_id]
1650
raise RevisionNotPresent(version_id, self)
1652
def check_versions_present(self, version_ids):
1653
"""Check that all specified versions are present."""
1655
for version_id in version_ids:
1656
if version_id not in cache:
1657
raise RevisionNotPresent(version_id, self._filename)
1660
class KnitGraphIndex(object):
1661
"""A knit index that builds on GraphIndex."""
1663
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1664
"""Construct a KnitGraphIndex on a graph_index.
1666
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1667
:param deltas: Allow delta-compressed records.
1668
:param add_callback: If not None, allow additions to the index and call
1669
this callback with a list of added GraphIndex nodes:
1670
[(node, value, node_refs), ...]
1671
:param parents: If True, record knits parents, if not do not record
1674
self._graph_index = graph_index
1675
self._deltas = deltas
1676
self._add_callback = add_callback
1677
self._parents = parents
1678
if deltas and not parents:
1679
raise KnitCorrupt(self, "Cannot do delta compression without "
1682
def _get_entries(self, keys, check_present=False):
1683
"""Get the entries for keys.
1685
:param keys: An iterable of index keys, - 1-tuples.
1690
for node in self._graph_index.iter_entries(keys):
1692
found_keys.add(node[1])
1694
# adapt parentless index to the rest of the code.
1695
for node in self._graph_index.iter_entries(keys):
1696
yield node[0], node[1], node[2], ()
1697
found_keys.add(node[1])
1699
missing_keys = keys.difference(found_keys)
1701
raise RevisionNotPresent(missing_keys.pop(), self)
1703
def _present_keys(self, version_ids):
1705
node[1] for node in self._get_entries(version_ids)])
1707
def _parentless_ancestry(self, versions):
1708
"""Honour the get_ancestry API for parentless knit indices."""
1709
wanted_keys = self._version_ids_to_keys(versions)
1710
present_keys = self._present_keys(wanted_keys)
1711
missing = set(wanted_keys).difference(present_keys)
1713
raise RevisionNotPresent(missing.pop(), self)
1714
return list(self._keys_to_version_ids(present_keys))
1716
def get_ancestry(self, versions, topo_sorted=True):
1717
"""See VersionedFile.get_ancestry."""
1718
if not self._parents:
1719
return self._parentless_ancestry(versions)
1720
# XXX: This will do len(history) index calls - perhaps
1721
# it should be altered to be a index core feature?
1722
# get a graph of all the mentioned versions:
1725
versions = self._version_ids_to_keys(versions)
1726
pending = set(versions)
1728
# get all pending nodes
1729
this_iteration = pending
1730
new_nodes = self._get_entries(this_iteration)
1733
for (index, key, value, node_refs) in new_nodes:
1734
# dont ask for ghosties - otherwise
1735
# we we can end up looping with pending
1736
# being entirely ghosted.
1737
graph[key] = [parent for parent in node_refs[0]
1738
if parent not in ghosts]
1740
for parent in graph[key]:
1741
# dont examine known nodes again
1746
ghosts.update(this_iteration.difference(found))
1747
if versions.difference(graph):
1748
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1750
result_keys = topo_sort(graph.items())
1752
result_keys = graph.iterkeys()
1753
return [key[0] for key in result_keys]
1755
def get_ancestry_with_ghosts(self, versions):
1756
"""See VersionedFile.get_ancestry."""
1757
if not self._parents:
1758
return self._parentless_ancestry(versions)
1759
# XXX: This will do len(history) index calls - perhaps
1760
# it should be altered to be a index core feature?
1761
# get a graph of all the mentioned versions:
1763
versions = self._version_ids_to_keys(versions)
1764
pending = set(versions)
1766
# get all pending nodes
1767
this_iteration = pending
1768
new_nodes = self._get_entries(this_iteration)
1770
for (index, key, value, node_refs) in new_nodes:
1771
graph[key] = node_refs[0]
1773
for parent in graph[key]:
1774
# dont examine known nodes again
1778
missing_versions = this_iteration.difference(graph)
1779
missing_needed = versions.intersection(missing_versions)
1781
raise RevisionNotPresent(missing_needed.pop(), self)
1782
for missing_version in missing_versions:
1783
# add a key, no parents
1784
graph[missing_version] = []
1785
pending.discard(missing_version) # don't look for it
1786
result_keys = topo_sort(graph.items())
1787
return [key[0] for key in result_keys]
1789
def get_build_details(self, version_ids):
1790
"""Get the method, index_memo and compression parent for version_ids.
1792
Ghosts are omitted from the result.
1794
:param version_ids: An iterable of version_ids.
1795
:return: A dict of version_id:(index_memo, compression_parent,
1796
parents, record_details).
1798
opaque structure to pass to read_records to extract the raw
1801
Content that this record is built upon, may be None
1803
Logical parents of this node
1805
extra information about the content which needs to be passed to
1806
Factory.parse_record
1809
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
1810
for entry in entries:
1811
version_id = self._keys_to_version_ids((entry[1],))[0]
1812
if not self._parents:
1815
parents = self._keys_to_version_ids(entry[3][0])
1816
if not self._deltas:
1817
compression_parent = None
1819
compression_parent_key = self._compression_parent(entry)
1820
if compression_parent_key:
1821
compression_parent = self._keys_to_version_ids(
1822
(compression_parent_key,))[0]
1824
compression_parent = None
1825
noeol = (entry[2][0] == 'N')
1826
if compression_parent:
1827
method = 'line-delta'
1830
result[version_id] = (self._node_to_position(entry),
1831
compression_parent, parents,
1835
def _compression_parent(self, an_entry):
1836
# return the key that an_entry is compressed against, or None
1837
# Grab the second parent list (as deltas implies parents currently)
1838
compression_parents = an_entry[3][1]
1839
if not compression_parents:
1841
assert len(compression_parents) == 1
1842
return compression_parents[0]
1844
def _get_method(self, node):
1845
if not self._deltas:
1847
if self._compression_parent(node):
1852
def get_graph(self):
1853
"""Return a list of the node:parents lists from this knit index."""
1854
if not self._parents:
1855
return [(key, ()) for key in self.get_versions()]
1857
for index, key, value, refs in self._graph_index.iter_all_entries():
1858
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1861
def iter_parents(self, version_ids):
1862
"""Iterate through the parents for many version ids.
1864
:param version_ids: An iterable yielding version_ids.
1865
:return: An iterator that yields (version_id, parents). Requested
1866
version_ids not present in the versioned file are simply skipped.
1867
The order is undefined, allowing for different optimisations in
1868
the underlying implementation.
1871
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1873
present_parents = set()
1874
for node in all_nodes:
1875
all_parents.update(node[3][0])
1876
# any node we are querying must be present
1877
present_parents.add(node[1])
1878
unknown_parents = all_parents.difference(present_parents)
1879
present_parents.update(self._present_keys(unknown_parents))
1880
for node in all_nodes:
1882
for parent in node[3][0]:
1883
if parent in present_parents:
1884
parents.append(parent[0])
1885
yield node[1][0], tuple(parents)
1887
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1888
yield node[1][0], ()
1890
def num_versions(self):
1891
return len(list(self._graph_index.iter_all_entries()))
1893
__len__ = num_versions
1895
def get_versions(self):
1896
"""Get all the versions in the file. not topologically sorted."""
1897
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1899
def has_version(self, version_id):
1900
"""True if the version is in the index."""
1901
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1903
def _keys_to_version_ids(self, keys):
1904
return tuple(key[0] for key in keys)
1906
def get_position(self, version_id):
1907
"""Return details needed to access the version.
1909
:return: a tuple (index, data position, size) to hand to the access
1910
logic to get the record.
1912
node = self._get_node(version_id)
1913
return self._node_to_position(node)
1915
def _node_to_position(self, node):
1916
"""Convert an index value to position details."""
1917
bits = node[2][1:].split(' ')
1918
return node[0], int(bits[0]), int(bits[1])
1920
def get_method(self, version_id):
1921
"""Return compression method of specified version."""
1922
return self._get_method(self._get_node(version_id))
1924
def _get_node(self, version_id):
1926
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1928
raise RevisionNotPresent(version_id, self)
1930
def get_options(self, version_id):
1931
"""Return a list representing options.
1935
node = self._get_node(version_id)
1936
options = [self._get_method(node)]
1937
if node[2][0] == 'N':
1938
options.append('no-eol')
1941
def get_parent_map(self, version_ids):
1942
"""Passed through to by KnitVersionedFile.get_parent_map."""
1943
nodes = self._get_entries(self._version_ids_to_keys(version_ids))
1947
result[node[1][0]] = self._keys_to_version_ids(node[3][0])
1950
result[node[1][0]] = ()
1953
def get_parents_with_ghosts(self, version_id):
1954
"""Return parents of specified version with ghosts."""
1956
return self.get_parent_map([version_id])[version_id]
1958
raise RevisionNotPresent(version_id, self)
1960
def check_versions_present(self, version_ids):
1961
"""Check that all specified versions are present."""
1962
keys = self._version_ids_to_keys(version_ids)
1963
present = self._present_keys(keys)
1964
missing = keys.difference(present)
1966
raise RevisionNotPresent(missing.pop(), self)
1968
def add_version(self, version_id, options, access_memo, parents):
1969
"""Add a version record to the index."""
1970
return self.add_versions(((version_id, options, access_memo, parents),))
1972
def add_versions(self, versions, random_id=False):
1973
"""Add multiple versions to the index.
1975
This function does not insert data into the Immutable GraphIndex
1976
backing the KnitGraphIndex, instead it prepares data for insertion by
1977
the caller and checks that it is safe to insert then calls
1978
self._add_callback with the prepared GraphIndex nodes.
1980
:param versions: a list of tuples:
1981
(version_id, options, pos, size, parents).
1982
:param random_id: If True the ids being added were randomly generated
1983
and no check for existence will be performed.
1985
if not self._add_callback:
1986
raise errors.ReadOnlyError(self)
1987
# we hope there are no repositories with inconsistent parentage
1992
for (version_id, options, access_memo, parents) in versions:
1993
index, pos, size = access_memo
1994
key = (version_id, )
1995
parents = tuple((parent, ) for parent in parents)
1996
if 'no-eol' in options:
2000
value += "%d %d" % (pos, size)
2001
if not self._deltas:
2002
if 'line-delta' in options:
2003
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2006
if 'line-delta' in options:
2007
node_refs = (parents, (parents[0],))
2009
node_refs = (parents, ())
2011
node_refs = (parents, )
2014
raise KnitCorrupt(self, "attempt to add node with parents "
2015
"in parentless index.")
2017
keys[key] = (value, node_refs)
2019
present_nodes = self._get_entries(keys)
2020
for (index, key, value, node_refs) in present_nodes:
2021
if (value, node_refs) != keys[key]:
2022
raise KnitCorrupt(self, "inconsistent details in add_versions"
2023
": %s %s" % ((value, node_refs), keys[key]))
2027
for key, (value, node_refs) in keys.iteritems():
2028
result.append((key, value, node_refs))
2030
for key, (value, node_refs) in keys.iteritems():
2031
result.append((key, value))
2032
self._add_callback(result)
2034
def _version_ids_to_keys(self, version_ids):
2035
return set((version_id, ) for version_id in version_ids)
2038
class _KnitAccess(object):
2039
"""Access to knit records in a .knit file."""
2041
def __init__(self, transport, filename, _file_mode, _dir_mode,
2042
_need_to_create, _create_parent_dir):
2043
"""Create a _KnitAccess for accessing and inserting data.
2045
:param transport: The transport the .knit is located on.
2046
:param filename: The filename of the .knit.
2048
self._transport = transport
2049
self._filename = filename
2050
self._file_mode = _file_mode
2051
self._dir_mode = _dir_mode
2052
self._need_to_create = _need_to_create
2053
self._create_parent_dir = _create_parent_dir
2055
def add_raw_records(self, sizes, raw_data):
2056
"""Add raw knit bytes to a storage area.
2058
The data is spooled to whereever the access method is storing data.
2060
:param sizes: An iterable containing the size of each raw data segment.
2061
:param raw_data: A bytestring containing the data.
2062
:return: A list of memos to retrieve the record later. Each memo is a
2063
tuple - (index, pos, length), where the index field is always None
2064
for the .knit access method.
2066
assert type(raw_data) == str, \
2067
'data must be plain bytes was %s' % type(raw_data)
2068
if not self._need_to_create:
2069
base = self._transport.append_bytes(self._filename, raw_data)
2071
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2072
create_parent_dir=self._create_parent_dir,
2073
mode=self._file_mode,
2074
dir_mode=self._dir_mode)
2075
self._need_to_create = False
2079
result.append((None, base, size))
2084
"""IFF this data access has its own storage area, initialise it.
2088
self._transport.put_bytes_non_atomic(self._filename, '',
2089
mode=self._file_mode)
2091
def open_file(self):
2092
"""IFF this data access can be represented as a single file, open it.
2094
For knits that are not mapped to a single file on disk this will
2097
:return: None or a file handle.
2100
return self._transport.get(self._filename)
2105
def get_raw_records(self, memos_for_retrieval):
2106
"""Get the raw bytes for a records.
2108
:param memos_for_retrieval: An iterable containing the (index, pos,
2109
length) memo for retrieving the bytes. The .knit method ignores
2110
the index as there is always only a single file.
2111
:return: An iterator over the bytes of the records.
2113
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2114
for pos, data in self._transport.readv(self._filename, read_vector):
2118
class _PackAccess(object):
2119
"""Access to knit records via a collection of packs."""
2121
def __init__(self, index_to_packs, writer=None):
2122
"""Create a _PackAccess object.
2124
:param index_to_packs: A dict mapping index objects to the transport
2125
and file names for obtaining data.
2126
:param writer: A tuple (pack.ContainerWriter, write_index) which
2127
contains the pack to write, and the index that reads from it will
2131
self.container_writer = writer[0]
2132
self.write_index = writer[1]
2134
self.container_writer = None
2135
self.write_index = None
2136
self.indices = index_to_packs
2138
def add_raw_records(self, sizes, raw_data):
2139
"""Add raw knit bytes to a storage area.
2141
The data is spooled to the container writer in one bytes-record per
2144
:param sizes: An iterable containing the size of each raw data segment.
2145
:param raw_data: A bytestring containing the data.
2146
:return: A list of memos to retrieve the record later. Each memo is a
2147
tuple - (index, pos, length), where the index field is the
2148
write_index object supplied to the PackAccess object.
2150
assert type(raw_data) == str, \
2151
'data must be plain bytes was %s' % type(raw_data)
2155
p_offset, p_length = self.container_writer.add_bytes_record(
2156
raw_data[offset:offset+size], [])
2158
result.append((self.write_index, p_offset, p_length))
2162
"""Pack based knits do not get individually created."""
2164
def get_raw_records(self, memos_for_retrieval):
2165
"""Get the raw bytes for a records.
2167
:param memos_for_retrieval: An iterable containing the (index, pos,
2168
length) memo for retrieving the bytes. The Pack access method
2169
looks up the pack to use for a given record in its index_to_pack
2171
:return: An iterator over the bytes of the records.
2173
# first pass, group into same-index requests
2175
current_index = None
2176
for (index, offset, length) in memos_for_retrieval:
2177
if current_index == index:
2178
current_list.append((offset, length))
2180
if current_index is not None:
2181
request_lists.append((current_index, current_list))
2182
current_index = index
2183
current_list = [(offset, length)]
2184
# handle the last entry
2185
if current_index is not None:
2186
request_lists.append((current_index, current_list))
2187
for index, offsets in request_lists:
2188
transport, path = self.indices[index]
2189
reader = pack.make_readv_reader(transport, path, offsets)
2190
for names, read_func in reader.iter_records():
2191
yield read_func(None)
2193
def open_file(self):
2194
"""Pack based knits have no single file."""
2197
def set_writer(self, writer, index, (transport, packname)):
2198
"""Set a writer to use for adding data."""
2199
if index is not None:
2200
self.indices[index] = (transport, packname)
2201
self.container_writer = writer
2202
self.write_index = index
2205
class _StreamAccess(object):
2206
"""A Knit Access object that provides data from a datastream.
2208
It also provides a fallback to present as unannotated data, annotated data
2209
from a *backing* access object.
2211
This is triggered by a index_memo which is pointing to a different index
2212
than this was constructed with, and is used to allow extracting full
2213
unannotated texts for insertion into annotated knits.
2216
def __init__(self, reader_callable, stream_index, backing_knit,
2218
"""Create a _StreamAccess object.
2220
:param reader_callable: The reader_callable from the datastream.
2221
This is called to buffer all the data immediately, for
2223
:param stream_index: The index the data stream this provides access to
2224
which will be present in native index_memo's.
2225
:param backing_knit: The knit object that will provide access to
2226
annotated texts which are not available in the stream, so as to
2227
create unannotated texts.
2228
:param orig_factory: The original content factory used to generate the
2229
stream. This is used for checking whether the thunk code for
2230
supporting _copy_texts will generate the correct form of data.
2232
self.data = reader_callable(None)
2233
self.stream_index = stream_index
2234
self.backing_knit = backing_knit
2235
self.orig_factory = orig_factory
2237
def get_raw_records(self, memos_for_retrieval):
2238
"""Get the raw bytes for a records.
2240
:param memos_for_retrieval: An iterable containing the (thunk_flag,
2241
index, start, end) memo for retrieving the bytes.
2242
:return: An iterator over the bytes of the records.
2244
# use a generator for memory friendliness
2245
for thunk_flag, version_id, start, end in memos_for_retrieval:
2246
if version_id is self.stream_index:
2247
yield self.data[start:end]
2249
# we have been asked to thunk. This thunking only occurs when
2250
# we are obtaining plain texts from an annotated backing knit
2251
# so that _copy_texts will work.
2252
# We could improve performance here by scanning for where we need
2253
# to do this and using get_line_list, then interleaving the output
2254
# as desired. However, for now, this is sufficient.
2255
if self.orig_factory.__class__ != KnitPlainFactory:
2256
raise errors.KnitCorrupt(
2257
self, 'Bad thunk request %r' % version_id)
2258
lines = self.backing_knit.get_lines(version_id)
2259
line_bytes = ''.join(lines)
2260
digest = sha_string(line_bytes)
2262
if lines[-1][-1] != '\n':
2263
lines[-1] = lines[-1] + '\n'
2265
orig_options = list(self.backing_knit._index.get_options(version_id))
2266
if 'fulltext' not in orig_options:
2267
if 'line-delta' not in orig_options:
2268
raise errors.KnitCorrupt(self,
2269
'Unknown compression method %r' % orig_options)
2270
orig_options.remove('line-delta')
2271
orig_options.append('fulltext')
2272
# We want plain data, because we expect to thunk only to allow text
2274
size, bytes = self.backing_knit._data._record_to_data(version_id,
2275
digest, lines, line_bytes)
2279
class _StreamIndex(object):
2280
"""A Knit Index object that uses the data map from a datastream."""
2282
def __init__(self, data_list, backing_index):
2283
"""Create a _StreamIndex object.
2285
:param data_list: The data_list from the datastream.
2286
:param backing_index: The index which will supply values for nodes
2287
referenced outside of this stream.
2289
self.data_list = data_list
2290
self.backing_index = backing_index
2291
self._by_version = {}
2293
for key, options, length, parents in data_list:
2294
self._by_version[key] = options, (pos, pos + length), parents
2297
def get_ancestry(self, versions, topo_sorted):
2298
"""Get an ancestry list for versions."""
2300
# Not needed for basic joins
2301
raise NotImplementedError(self.get_ancestry)
2302
# get a graph of all the mentioned versions:
2303
# Little ugly - basically copied from KnitIndex, but don't want to
2304
# accidentally incorporate too much of that index's code.
2306
pending = set(versions)
2307
cache = self._by_version
2309
version = pending.pop()
2312
parents = [p for p in cache[version][2] if p in cache]
2314
raise RevisionNotPresent(version, self)
2315
# if not completed and not a ghost
2316
pending.update([p for p in parents if p not in ancestry])
2317
ancestry.add(version)
2318
return list(ancestry)
2320
def get_build_details(self, version_ids):
2321
"""Get the method, index_memo and compression parent for version_ids.
2323
Ghosts are omitted from the result.
2325
:param version_ids: An iterable of version_ids.
2326
:return: A dict of version_id:(index_memo, compression_parent,
2327
parents, record_details).
2329
opaque structure to pass to read_records to extract the raw
2332
Content that this record is built upon, may be None
2334
Logical parents of this node
2336
extra information about the content which needs to be passed to
2337
Factory.parse_record
2340
for version_id in version_ids:
2342
method = self.get_method(version_id)
2343
except errors.RevisionNotPresent:
2344
# ghosts are omitted
2346
parent_ids = self.get_parents_with_ghosts(version_id)
2347
noeol = ('no-eol' in self.get_options(version_id))
2348
if method == 'fulltext':
2349
compression_parent = None
2351
compression_parent = parent_ids[0]
2352
index_memo = self.get_position(version_id)
2353
result[version_id] = (index_memo, compression_parent,
2354
parent_ids, (method, noeol))
2357
def get_method(self, version_id):
2358
"""Return compression method of specified version."""
2360
options = self._by_version[version_id][0]
2362
# Strictly speaking this should check in the backing knit, but
2363
# until we have a test to discriminate, this will do.
2364
return self.backing_index.get_method(version_id)
2365
if 'fulltext' in options:
2367
elif 'line-delta' in options:
2370
raise errors.KnitIndexUnknownMethod(self, options)
2372
def get_options(self, version_id):
2373
"""Return a list representing options.
2378
return self._by_version[version_id][0]
2380
return self.backing_index.get_options(version_id)
2382
def get_parent_map(self, version_ids):
2383
"""Passed through to by KnitVersionedFile.get_parent_map."""
2386
for version_id in version_ids:
2388
result[version_id] = self._by_version[version_id][2]
2390
pending_ids.add(version_id)
2391
result.update(self.backing_index.get_parent_map(pending_ids))
2394
def get_parents_with_ghosts(self, version_id):
2395
"""Return parents of specified version with ghosts."""
2397
return self.get_parent_map([version_id])[version_id]
2399
raise RevisionNotPresent(version_id, self)
2401
def get_position(self, version_id):
2402
"""Return details needed to access the version.
2404
_StreamAccess has the data as a big array, so we return slice
2405
coordinates into that (as index_memo's are opaque outside the
2406
index and matching access class).
2408
:return: a tuple (thunk_flag, index, start, end). If thunk_flag is
2409
False, index will be self, otherwise it will be a version id.
2412
start, end = self._by_version[version_id][1]
2413
return False, self, start, end
2415
# Signal to the access object to handle this from the backing knit.
2416
return (True, version_id, None, None)
2418
def get_versions(self):
2419
"""Get all the versions in the stream."""
2420
return self._by_version.keys()
2422
def iter_parents(self, version_ids):
2423
"""Iterate through the parents for many version ids.
2425
:param version_ids: An iterable yielding version_ids.
2426
:return: An iterator that yields (version_id, parents). Requested
2427
version_ids not present in the versioned file are simply skipped.
2428
The order is undefined, allowing for different optimisations in
2429
the underlying implementation.
2432
for version in version_ids:
2434
result.append((version, self._by_version[version][2]))
2440
class _KnitData(object):
2441
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2443
The KnitData class provides the logic for parsing and using knit records,
2444
making use of an access method for the low level read and write operations.
2447
def __init__(self, access):
2448
"""Create a KnitData object.
2450
:param access: The access method to use. Access methods such as
2451
_KnitAccess manage the insertion of raw records and the subsequent
2452
retrieval of the same.
2454
self._access = access
2455
self._checked = False
2456
# TODO: jam 20060713 conceptually, this could spill to disk
2457
# if the cached size gets larger than a certain amount
2458
# but it complicates the model a bit, so for now just use
2459
# a simple dictionary
2461
self._do_cache = False
2463
def enable_cache(self):
2464
"""Enable caching of reads."""
2465
self._do_cache = True
2467
def clear_cache(self):
2468
"""Clear the record cache."""
2469
self._do_cache = False
2472
def _open_file(self):
2473
return self._access.open_file()
2475
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2476
"""Convert version_id, digest, lines into a raw data block.
2478
:param dense_lines: The bytes of lines but in a denser form. For
2479
instance, if lines is a list of 1000 bytestrings each ending in \n,
2480
dense_lines may be a list with one line in it, containing all the
2481
1000's lines and their \n's. Using dense_lines if it is already
2482
known is a win because the string join to create bytes in this
2483
function spends less time resizing the final string.
2484
:return: (len, a StringIO instance with the raw data ready to read.)
2486
# Note: using a string copy here increases memory pressure with e.g.
2487
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2488
# when doing the initial commit of a mozilla tree. RBC 20070921
2489
bytes = ''.join(chain(
2490
["version %s %d %s\n" % (version_id,
2493
dense_lines or lines,
2494
["end %s\n" % version_id]))
2495
assert bytes.__class__ == str
2496
compressed_bytes = bytes_to_gzip(bytes)
2497
return len(compressed_bytes), compressed_bytes
2499
def add_raw_records(self, sizes, raw_data):
2500
"""Append a prepared record to the data file.
2502
:param sizes: An iterable containing the size of each raw data segment.
2503
:param raw_data: A bytestring containing the data.
2504
:return: a list of index data for the way the data was stored.
2505
See the access method add_raw_records documentation for more
2508
return self._access.add_raw_records(sizes, raw_data)
2510
def _parse_record_header(self, version_id, raw_data):
2511
"""Parse a record header for consistency.
2513
:return: the header and the decompressor stream.
2514
as (stream, header_record)
2516
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2518
rec = self._check_header(version_id, df.readline())
2519
except Exception, e:
2520
raise KnitCorrupt(self._access,
2521
"While reading {%s} got %s(%s)"
2522
% (version_id, e.__class__.__name__, str(e)))
2525
def _check_header(self, version_id, line):
2528
raise KnitCorrupt(self._access,
2529
'unexpected number of elements in record header')
2530
if rec[1] != version_id:
2531
raise KnitCorrupt(self._access,
2532
'unexpected version, wanted %r, got %r'
2533
% (version_id, rec[1]))
2536
def _parse_record(self, version_id, data):
2538
# 4168 calls in 2880 217 internal
2539
# 4168 calls to _parse_record_header in 2121
2540
# 4168 calls to readlines in 330
2541
df = GzipFile(mode='rb', fileobj=StringIO(data))
2544
record_contents = df.readlines()
2545
except Exception, e:
2546
raise KnitCorrupt(self._access,
2547
"While reading {%s} got %s(%s)"
2548
% (version_id, e.__class__.__name__, str(e)))
2549
header = record_contents.pop(0)
2550
rec = self._check_header(version_id, header)
2552
last_line = record_contents.pop()
2553
if len(record_contents) != int(rec[2]):
2554
raise KnitCorrupt(self._access,
2555
'incorrect number of lines %s != %s'
2557
% (len(record_contents), int(rec[2]),
2559
if last_line != 'end %s\n' % rec[1]:
2560
raise KnitCorrupt(self._access,
2561
'unexpected version end line %r, wanted %r'
2562
% (last_line, version_id))
2564
return record_contents, rec[3]
2566
def read_records_iter_raw(self, records):
2567
"""Read text records from data file and yield raw data.
2569
This unpacks enough of the text record to validate the id is
2570
as expected but thats all.
2572
# setup an iterator of the external records:
2573
# uses readv so nice and fast we hope.
2575
# grab the disk data needed.
2577
# Don't check _cache if it is empty
2578
needed_offsets = [index_memo for version_id, index_memo
2580
if version_id not in self._cache]
2582
needed_offsets = [index_memo for version_id, index_memo
2585
raw_records = self._access.get_raw_records(needed_offsets)
2587
for version_id, index_memo in records:
2588
if version_id in self._cache:
2589
# This data has already been validated
2590
data = self._cache[version_id]
2592
data = raw_records.next()
2594
self._cache[version_id] = data
2596
# validate the header
2597
df, rec = self._parse_record_header(version_id, data)
2599
yield version_id, data
2601
def read_records_iter(self, records):
2602
"""Read text records from data file and yield result.
2604
The result will be returned in whatever is the fastest to read.
2605
Not by the order requested. Also, multiple requests for the same
2606
record will only yield 1 response.
2607
:param records: A list of (version_id, pos, len) entries
2608
:return: Yields (version_id, contents, digest) in the order
2609
read, not the order requested
2615
# Skip records we have alread seen
2616
yielded_records = set()
2617
needed_records = set()
2618
for record in records:
2619
if record[0] in self._cache:
2620
if record[0] in yielded_records:
2622
yielded_records.add(record[0])
2623
data = self._cache[record[0]]
2624
content, digest = self._parse_record(record[0], data)
2625
yield (record[0], content, digest)
2627
needed_records.add(record)
2628
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2630
needed_records = sorted(set(records), key=operator.itemgetter(1))
2632
if not needed_records:
2635
# The transport optimizes the fetching as well
2636
# (ie, reads continuous ranges.)
2637
raw_data = self._access.get_raw_records(
2638
[index_memo for version_id, index_memo in needed_records])
2640
for (version_id, index_memo), data in \
2641
izip(iter(needed_records), raw_data):
2642
content, digest = self._parse_record(version_id, data)
2644
self._cache[version_id] = data
2645
yield version_id, content, digest
2647
def read_records(self, records):
2648
"""Read records into a dictionary."""
2650
for record_id, content, digest in \
2651
self.read_records_iter(records):
2652
components[record_id] = (content, digest)
2656
class InterKnit(InterVersionedFile):
2657
"""Optimised code paths for knit to knit operations."""
2659
_matching_file_from_factory = KnitVersionedFile
2660
_matching_file_to_factory = KnitVersionedFile
2663
def is_compatible(source, target):
2664
"""Be compatible with knits. """
2666
return (isinstance(source, KnitVersionedFile) and
2667
isinstance(target, KnitVersionedFile))
2668
except AttributeError:
2671
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2672
"""Copy texts to the target by extracting and adding them one by one.
2674
see join() for the parameter definitions.
2676
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2677
# --- the below is factorable out with VersionedFile.join, but wait for
2678
# VersionedFiles, it may all be simpler then.
2679
graph = Graph(self.source)
2680
search = graph._make_breadth_first_searcher(version_ids)
2681
transitive_ids = set()
2682
map(transitive_ids.update, list(search))
2683
parent_map = self.source.get_parent_map(transitive_ids)
2684
order = topo_sort(parent_map.items())
2686
def size_of_content(content):
2687
return sum(len(line) for line in content.text())
2688
# Cache at most 10MB of parent texts
2689
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2690
compute_size=size_of_content)
2691
# TODO: jam 20071116 It would be nice to have a streaming interface to
2692
# get multiple texts from a source. The source could be smarter
2693
# about how it handled intermediate stages.
2694
# get_line_list() or make_mpdiffs() seem like a possibility, but
2695
# at the moment they extract all full texts into memory, which
2696
# causes us to store more than our 3x fulltext goal.
2697
# Repository.iter_files_bytes() may be another possibility
2698
to_process = [version for version in order
2699
if version not in self.target]
2700
total = len(to_process)
2701
pb = ui.ui_factory.nested_progress_bar()
2703
for index, version in enumerate(to_process):
2704
pb.update('Converting versioned data', index, total)
2705
sha1, num_bytes, parent_text = self.target.add_lines(version,
2706
self.source.get_parents_with_ghosts(version),
2707
self.source.get_lines(version),
2708
parent_texts=parent_cache)
2709
parent_cache[version] = parent_text
2714
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2715
"""See InterVersionedFile.join."""
2716
assert isinstance(self.source, KnitVersionedFile)
2717
assert isinstance(self.target, KnitVersionedFile)
2719
# If the source and target are mismatched w.r.t. annotations vs
2720
# plain, the data needs to be converted accordingly
2721
if self.source.factory.annotated == self.target.factory.annotated:
2723
elif self.source.factory.annotated:
2724
converter = self._anno_to_plain_converter
2726
# We're converting from a plain to an annotated knit. Copy them
2727
# across by full texts.
2728
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2730
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2734
pb = ui.ui_factory.nested_progress_bar()
2736
version_ids = list(version_ids)
2737
if None in version_ids:
2738
version_ids.remove(None)
2740
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2742
this_versions = set(self.target._index.get_versions())
2743
# XXX: For efficiency we should not look at the whole index,
2744
# we only need to consider the referenced revisions - they
2745
# must all be present, or the method must be full-text.
2746
# TODO, RBC 20070919
2747
needed_versions = self.source_ancestry - this_versions
2749
if not needed_versions:
2751
full_list = topo_sort(
2752
self.source.get_parent_map(self.source.versions()))
2754
version_list = [i for i in full_list if (not self.target.has_version(i)
2755
and i in needed_versions)]
2759
copy_queue_records = []
2761
for version_id in version_list:
2762
options = self.source._index.get_options(version_id)
2763
parents = self.source._index.get_parents_with_ghosts(version_id)
2764
# check that its will be a consistent copy:
2765
for parent in parents:
2766
# if source has the parent, we must :
2767
# * already have it or
2768
# * have it scheduled already
2769
# otherwise we don't care
2770
assert (self.target.has_version(parent) or
2771
parent in copy_set or
2772
not self.source.has_version(parent))
2773
index_memo = self.source._index.get_position(version_id)
2774
copy_queue_records.append((version_id, index_memo))
2775
copy_queue.append((version_id, options, parents))
2776
copy_set.add(version_id)
2778
# data suck the join:
2780
total = len(version_list)
2783
for (version_id, raw_data), \
2784
(version_id2, options, parents) in \
2785
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2787
assert version_id == version_id2, 'logic error, inconsistent results'
2789
pb.update("Joining knit", count, total)
2791
size, raw_data = converter(raw_data, version_id, options,
2794
size = len(raw_data)
2795
raw_records.append((version_id, options, parents, size))
2796
raw_datum.append(raw_data)
2797
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2802
def _anno_to_plain_converter(self, raw_data, version_id, options,
2804
"""Convert annotated content to plain content."""
2805
data, digest = self.source._data._parse_record(version_id, raw_data)
2806
if 'fulltext' in options:
2807
content = self.source.factory.parse_fulltext(data, version_id)
2808
lines = self.target.factory.lower_fulltext(content)
2810
delta = self.source.factory.parse_line_delta(data, version_id,
2812
lines = self.target.factory.lower_line_delta(delta)
2813
return self.target._data._record_to_data(version_id, digest, lines)
2816
InterVersionedFile.register_optimiser(InterKnit)
2819
class WeaveToKnit(InterVersionedFile):
2820
"""Optimised code paths for weave to knit operations."""
2822
_matching_file_from_factory = bzrlib.weave.WeaveFile
2823
_matching_file_to_factory = KnitVersionedFile
2826
def is_compatible(source, target):
2827
"""Be compatible with weaves to knits."""
2829
return (isinstance(source, bzrlib.weave.Weave) and
2830
isinstance(target, KnitVersionedFile))
2831
except AttributeError:
2834
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2835
"""See InterVersionedFile.join."""
2836
assert isinstance(self.source, bzrlib.weave.Weave)
2837
assert isinstance(self.target, KnitVersionedFile)
2839
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2844
pb = ui.ui_factory.nested_progress_bar()
2846
version_ids = list(version_ids)
2848
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2849
this_versions = set(self.target._index.get_versions())
2850
needed_versions = self.source_ancestry - this_versions
2852
if not needed_versions:
2854
full_list = topo_sort(
2855
self.source.get_parent_map(self.source.versions()))
2857
version_list = [i for i in full_list if (not self.target.has_version(i)
2858
and i in needed_versions)]
2862
total = len(version_list)
2863
parent_map = self.source.get_parent_map(version_list)
2864
for version_id in version_list:
2865
pb.update("Converting to knit", count, total)
2866
parents = parent_map[version_id]
2867
# check that its will be a consistent copy:
2868
for parent in parents:
2869
# if source has the parent, we must already have it
2870
assert (self.target.has_version(parent))
2871
self.target.add_lines(
2872
version_id, parents, self.source.get_lines(version_id))
2879
InterVersionedFile.register_optimiser(WeaveToKnit)
2882
# Deprecated, use PatienceSequenceMatcher instead
2883
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2886
def annotate_knit(knit, revision_id):
2887
"""Annotate a knit with no cached annotations.
2889
This implementation is for knits with no cached annotations.
2890
It will work for knits with cached annotations, but this is not
2893
annotator = _KnitAnnotator(knit)
2894
return iter(annotator.annotate(revision_id))
2897
class _KnitAnnotator(object):
2898
"""Build up the annotations for a text."""
2900
def __init__(self, knit):
2903
# Content objects, differs from fulltexts because of how final newlines
2904
# are treated by knits. the content objects here will always have a
2906
self._fulltext_contents = {}
2908
# Annotated lines of specific revisions
2909
self._annotated_lines = {}
2911
# Track the raw data for nodes that we could not process yet.
2912
# This maps the revision_id of the base to a list of children that will
2913
# annotated from it.
2914
self._pending_children = {}
2916
# Nodes which cannot be extracted
2917
self._ghosts = set()
2919
# Track how many children this node has, so we know if we need to keep
2921
self._annotate_children = {}
2922
self._compression_children = {}
2924
self._all_build_details = {}
2925
# The children => parent revision_id graph
2926
self._revision_id_graph = {}
2928
self._heads_provider = None
2930
self._nodes_to_keep_annotations = set()
2931
self._generations_until_keep = 100
2933
def set_generations_until_keep(self, value):
2934
"""Set the number of generations before caching a node.
2936
Setting this to -1 will cache every merge node, setting this higher
2937
will cache fewer nodes.
2939
self._generations_until_keep = value
2941
def _add_fulltext_content(self, revision_id, content_obj):
2942
self._fulltext_contents[revision_id] = content_obj
2943
# TODO: jam 20080305 It might be good to check the sha1digest here
2944
return content_obj.text()
2946
def _check_parents(self, child, nodes_to_annotate):
2947
"""Check if all parents have been processed.
2949
:param child: A tuple of (rev_id, parents, raw_content)
2950
:param nodes_to_annotate: If child is ready, add it to
2951
nodes_to_annotate, otherwise put it back in self._pending_children
2953
for parent_id in child[1]:
2954
if (parent_id not in self._annotated_lines):
2955
# This parent is present, but another parent is missing
2956
self._pending_children.setdefault(parent_id,
2960
# This one is ready to be processed
2961
nodes_to_annotate.append(child)
2963
def _add_annotation(self, revision_id, fulltext, parent_ids,
2964
left_matching_blocks=None):
2965
"""Add an annotation entry.
2967
All parents should already have been annotated.
2968
:return: A list of children that now have their parents satisfied.
2970
a = self._annotated_lines
2971
annotated_parent_lines = [a[p] for p in parent_ids]
2972
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2973
fulltext, revision_id, left_matching_blocks,
2974
heads_provider=self._get_heads_provider()))
2975
self._annotated_lines[revision_id] = annotated_lines
2976
for p in parent_ids:
2977
ann_children = self._annotate_children[p]
2978
ann_children.remove(revision_id)
2979
if (not ann_children
2980
and p not in self._nodes_to_keep_annotations):
2981
del self._annotated_lines[p]
2982
del self._all_build_details[p]
2983
if p in self._fulltext_contents:
2984
del self._fulltext_contents[p]
2985
# Now that we've added this one, see if there are any pending
2986
# deltas to be done, certainly this parent is finished
2987
nodes_to_annotate = []
2988
for child in self._pending_children.pop(revision_id, []):
2989
self._check_parents(child, nodes_to_annotate)
2990
return nodes_to_annotate
2992
def _get_build_graph(self, revision_id):
2993
"""Get the graphs for building texts and annotations.
2995
The data you need for creating a full text may be different than the
2996
data you need to annotate that text. (At a minimum, you need both
2997
parents to create an annotation, but only need 1 parent to generate the
3000
:return: A list of (revision_id, index_memo) records, suitable for
3001
passing to read_records_iter to start reading in the raw data fro/
3004
if revision_id in self._annotated_lines:
3007
pending = set([revision_id])
3012
# get all pending nodes
3014
this_iteration = pending
3015
build_details = self._knit._index.get_build_details(this_iteration)
3016
self._all_build_details.update(build_details)
3017
# new_nodes = self._knit._index._get_entries(this_iteration)
3019
for rev_id, details in build_details.iteritems():
3020
(index_memo, compression_parent, parents,
3021
record_details) = details
3022
self._revision_id_graph[rev_id] = parents
3023
records.append((rev_id, index_memo))
3024
# Do we actually need to check _annotated_lines?
3025
pending.update(p for p in parents
3026
if p not in self._all_build_details)
3027
if compression_parent:
3028
self._compression_children.setdefault(compression_parent,
3031
for parent in parents:
3032
self._annotate_children.setdefault(parent,
3034
num_gens = generation - kept_generation
3035
if ((num_gens >= self._generations_until_keep)
3036
and len(parents) > 1):
3037
kept_generation = generation
3038
self._nodes_to_keep_annotations.add(rev_id)
3040
missing_versions = this_iteration.difference(build_details.keys())
3041
self._ghosts.update(missing_versions)
3042
for missing_version in missing_versions:
3043
# add a key, no parents
3044
self._revision_id_graph[missing_version] = ()
3045
pending.discard(missing_version) # don't look for it
3046
# XXX: This should probably be a real exception, as it is a data
3048
assert not self._ghosts.intersection(self._compression_children), \
3049
"We cannot have nodes which have a compression parent of a ghost."
3050
# Cleanout anything that depends on a ghost so that we don't wait for
3051
# the ghost to show up
3052
for node in self._ghosts:
3053
if node in self._annotate_children:
3054
# We won't be building this node
3055
del self._annotate_children[node]
3056
# Generally we will want to read the records in reverse order, because
3057
# we find the parent nodes after the children
3061
def _annotate_records(self, records):
3062
"""Build the annotations for the listed records."""
3063
# We iterate in the order read, rather than a strict order requested
3064
# However, process what we can, and put off to the side things that
3065
# still need parents, cleaning them up when those parents are
3067
for (rev_id, record,
3068
digest) in self._knit._data.read_records_iter(records):
3069
if rev_id in self._annotated_lines:
3071
parent_ids = self._revision_id_graph[rev_id]
3072
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3073
details = self._all_build_details[rev_id]
3074
(index_memo, compression_parent, parents,
3075
record_details) = details
3076
nodes_to_annotate = []
3077
# TODO: Remove the punning between compression parents, and
3078
# parent_ids, we should be able to do this without assuming
3080
if len(parent_ids) == 0:
3081
# There are no parents for this node, so just add it
3082
# TODO: This probably needs to be decoupled
3083
assert compression_parent is None
3084
fulltext_content, delta = self._knit.factory.parse_record(
3085
rev_id, record, record_details, None)
3086
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3087
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3088
parent_ids, left_matching_blocks=None))
3090
child = (rev_id, parent_ids, record)
3091
# Check if all the parents are present
3092
self._check_parents(child, nodes_to_annotate)
3093
while nodes_to_annotate:
3094
# Should we use a queue here instead of a stack?
3095
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3096
(index_memo, compression_parent, parents,
3097
record_details) = self._all_build_details[rev_id]
3098
if compression_parent is not None:
3099
comp_children = self._compression_children[compression_parent]
3100
assert rev_id in comp_children
3101
# If there is only 1 child, it is safe to reuse this
3103
reuse_content = (len(comp_children) == 1
3104
and compression_parent not in
3105
self._nodes_to_keep_annotations)
3107
# Remove it from the cache since it will be changing
3108
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3109
# Make sure to copy the fulltext since it might be
3111
parent_fulltext = list(parent_fulltext_content.text())
3113
parent_fulltext_content = self._fulltext_contents[compression_parent]
3114
parent_fulltext = parent_fulltext_content.text()
3115
comp_children.remove(rev_id)
3116
fulltext_content, delta = self._knit.factory.parse_record(
3117
rev_id, record, record_details,
3118
parent_fulltext_content,
3119
copy_base_content=(not reuse_content))
3120
fulltext = self._add_fulltext_content(rev_id,
3122
blocks = KnitContent.get_line_delta_blocks(delta,
3123
parent_fulltext, fulltext)
3125
fulltext_content = self._knit.factory.parse_fulltext(
3127
fulltext = self._add_fulltext_content(rev_id,
3130
nodes_to_annotate.extend(
3131
self._add_annotation(rev_id, fulltext, parent_ids,
3132
left_matching_blocks=blocks))
3134
def _get_heads_provider(self):
3135
"""Create a heads provider for resolving ancestry issues."""
3136
if self._heads_provider is not None:
3137
return self._heads_provider
3138
parent_provider = _mod_graph.DictParentsProvider(
3139
self._revision_id_graph)
3140
graph_obj = _mod_graph.Graph(parent_provider)
3141
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3142
self._heads_provider = head_cache
3145
def annotate(self, revision_id):
3146
"""Return the annotated fulltext at the given revision.
3148
:param revision_id: The revision id for this file
3150
records = self._get_build_graph(revision_id)
3151
if revision_id in self._ghosts:
3152
raise errors.RevisionNotPresent(revision_id, self._knit)
3153
self._annotate_records(records)
3154
return self._annotated_lines[revision_id]
3158
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3160
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
